Aperio IHC Membrane Image Analysis User`s Guide

Aperio IHC Membrane Image Analysis User`s Guide
IHC Membrane Image
Analysis
User’s Guide
©Copyright 2007 Aperio Technologies, Inc.
Part Number/Revision: MAN-0026, Revision B
Date: January 2, 2007
This document applies to software versions Release 8.0 and later.
All rights reserved. This document may not be copied in whole or in part or reproduced in any other media without the
express written permission of Aperio Technologies, Inc. Please note that under copyright law, copying includes translation
into another language.
User Resources
For the latest information on Aperio Technologies products and services, please visit the Aperio Technologies website at:
http://www.aperio.com.
Disclaimers
This manual is not a substitute for the detailed operator training provided by Aperio Technologies, Inc., or for other
advanced instruction. Aperio Technologies Field Representatives should be contacted immediately for assistance in the
event of any instrument malfunction. Installation of hardware should only be performed by a certified Aperio Technologies
Service Engineer.
ImageServer is intended for use with the svs file format (the native format for digital slides created by scanning glass slides
with the ScanScope scanner). Educators will use Aperio software to view and modify digital slides in Composite WebSlide
(CWS) format.
CAUTION
For research and education uses only, not for use in diagnostic procedures. This product has not been approved or cleared
as a medical device by the U.S. Food and Drug Administration. The data and images obtained or viewed using this product
are not intended for clinical or diagnostic use.
Trademarks and Patents
ScanScope is a registered trademark and ImageServer, TMALab, ImageScope, and Spectrum are trademarks of Aperio
Technologies, Inc. All other trade names and trademarks are the property of their respective holders.
Aperio products are protected by U.S. Patents: 6,711,283; 6,917,696; 7,035,478; and 7,116,440; and licensed under one or more
of the following U.S. Patents: 6,101,265; 6,272,235; 6,522,774; 6,775,402; 6,396,941; 6,674,881; 6,226,392; 6,404,906; 6,674,884;
and 6,466,690.
Contact Information
Headquarters:
European Office:
Aperio Technologies, Inc.
1430 Vantage Court, Suite 106
Vista, CA 92081
United States
Aperio
3 The Sanctuary
Eden Office Park
Ham Green
Bristol BS20 0DD, UK
United States of America
Tel: 866-478-4111 (toll free)
Fax: 760-539-1116
Customer Service
Tel: 866-478-4111 (toll free)
Technical Support
Tel: 866-478-4111 (toll free)
Email: support@aperio.com
Europe
Tel: +44 (0) 1275 375123
Fax: +44(0) 1275 373501
Customer Service
Tel: +44 (0) 1275 375123
Technical Support
Tel: +44 (0) 1275 375123
Email: europesupport@aperio.com
ii
IHC Membrane Image Analysis User’s Guide
Contents
CHAPTER 1 - OVERVIEW ............................................................................................................... 1
IHC Workflow .......................................................................................................................................................1
Who This Guide Is Intended For........................................................................................................................2
Intended Use ..........................................................................................................................................................2
Algorithm Description .........................................................................................................................................2
CHAPTER 2 - ANALYSIS STEP-BY-STEP ............................................................................................. 5
Access a Digital Slide in SIM (Spectrum Information Manager) ................................................................5
Patient Data Confidentiality.............................................................................................................................5
Manual Read of a Slide ........................................................................................................................................6
Image Analysis in ImageScope...........................................................................................................................8
Outline Tumor Regions.....................................................................................................................................8
Run Algorithm ...................................................................................................................................................9
View Results .....................................................................................................................................................11
Final Score ............................................................................................................................................................13
More Information................................................................................................................................................13
CHAPTER 3 - INSTALLATION ........................................................................................................ 15
CHAPTER 4 - SLIDE PREPARATION ................................................................................................. 17
CHAPTER 5 - SCANNING ............................................................................................................. 19
CHAPTER 6 - ALGORITHM SETUP .................................................................................................. 21
Data Sets................................................................................................................................................................21
Cell Feature Detection Parameters ...................................................................................................................27
Scoring Scheme Parameters...............................................................................................................................34
Validate Algorithm Setup..................................................................................................................................37
CHAPTER 7 - VALIDATION .......................................................................................................... 39
IHC Membrane Image Analysis User’s Guide
iii
Contents
iv
IHC Membrane Image Analysis User’s Guide
1
Overview
This chapter provides an overview of Aperio’s IHC (ImmunoHistoChemistry)
Membrane Image Analysis algorithm.
The following chapters of this manual provide step-by-step instructions for a
pathologist on how to use the algorithm as well as detailed information on
proper slide preparation and scanning, and the installation, setup and
performance validation of the algorithm.
IHC Workflow
A lab receives a specimen from a biopsy and prepares a block from which an H&E (Haematoxylin and
Eosin) slide is prepared. The typical IHC (ImmunoHistoChemistry) workflow starts with the diagnosis of
cancer by a pathologist reviewing the H&E slide.
Once the pathologist has determined that a patient has cancer he/she orders an IHC panel. An IHC panel
consists of consecutively cut slides from the same specimen block that are then stained with different IHC
stains. A typical IHC panel for breast cancer includes the following IHC stains: HER2 (Human Epidermal
growth factor Receptor 2), ER (Estrogen Receptor) and PR (Progesterone Receptor). The pathologist then
provides a semi-quantitative assessment of the IHC slides to determine the appropriate treatment option
(prognostic).
HER2 is a membrane stain from which the pathologist assesses the intensity and completeness of the
membrane staining of the tumor cells and determines a score of 0, 1+, 2+ or 3+ for the slide. HER2 is used
specifically to assess Herceptin® (Trastuzumab) as a treatment option and to determine the
aggressiveness of breast tumors.
ER and PR are nuclear stains from which the pathologist assesses the staining intensity of the nuclei of
the tumor cells and determines the percentage of positive stained nuclei as a percentage between 0 to
100% and the average staining intensity of the positive nuclei as a score of 0, 1+, 2+ or 3+. Different labs
use different scoring schemes for the nuclear stains. For example, some labs use only the percentage of
positive nuclei as a score and use different cut-off thresholds of 1%, 5% or 10% for the interpretation;
other labs use the percentage of positive nuclei and the average staining intensity of the positive nuclei to
calculate a combined score like the Allred Score.
IHC Membrane Image Analysis User’s Guide
1
Chapter 1 – Overview
Who This Guide Is Intended For
This guide is a reference document that covers all aspects of using the Aperio IHC
(ImmunoHistoChemistry) Membrane Image Analysis algorithm. Different steps of your image analysis
workflow may involve different people with different skills and responsibilities. Therefore, the chapters
in this guide are organized by function so that a particular reader can go directly to the section of interest
to him or her:
Functional Area
Responsible Person
Where to Find Information
Overview
All readers
Chapter 1, “Overview” on page 1
Step-by step analysis—everything
Pathologist
Chapter 2, “Analysis Step-by-Step,” on page 5
Lab manager
Chapter 4, “Slide Preparation,” on page 17
ScanScope operator
Chapter 5, “Scanning,” on page 19
Installation of the algorithm
IT administrator
Chapter 2, “Installation,” on page 5
How to set up the parameters of
Person responsible for
Chapter 6, “Algorithm Setup,” on page 21
the algorithm to optimize analysis
image analysis services
the pathologist needs to know to
perform daily tasks
Tips for ensuring your glass slides
are optimized for image analysis
Tips on performing high quality
slide scans
results
Validation of algorithm
Medical Director
Chapter 7, “Validation,” on page 39
performance
Intended Use
The IHC (ImmunoHistoChemistry) Membrane Image Analysis algorithm is intended to be used as an aid
to the pathologists for the assessment of IHC HER2 (Human Epidermal growth factor Receptor 2) stained
breast tissue slides.
Although the IHC Membrane Image Analysis algorithm has been optimized for IHC HER2 staining on
breast tissue, it is a general membrane algorithm that can be used for other tissue types and membrane
stains.
CAUTION:
For research and educational uses only, not for use in diagnostic procedures. This product has not been
approved or cleared as a medical device by the U.S. Food and Drug Administration. The data and images
obtained or viewed using this product are not intended for clinical or diagnostic use.
Algorithm Description
Prior to running the algorithm, a qualified pathologist needs to use the ImageScope annotation tools to
outline a set of tumor-cell only regions that are representative of the tumor.
The IHC Membrane Image Analysis algorithm detects the membrane staining for the individual tumor
cells in the selected regions and quantifies the intensity and completeness of the membrane staining.
Tumor cells are individually classified as 0, 1+, 2+ and 3+ based on their membrane staining intensity and
completeness. A tumor cell is classified 1+ when there is only partial membrane staining or weak
2
IHC Membrane Image Analysis User’s Guide
Chapter 1 – Overview
membrane staining. A tumor cell is classified 2+ when there is moderate and complete membrane
staining. A tumor cell is classified 3+ when there is intense and complete membrane staining.
Based on the percentages of 0, 1+, 2+ and 3+ classified cell scores, a slide score of 0, 1+, 2+ or 3+ is
determined. The score is 3+ if more than 10% of the cells are stained at the 3+ level. The score is 2+ if more
than 10% of the cells are stained at the 2+ and higher levels. The score is 1+ if more than 10% of the cells
are stained at the 1+ and higher level. The score is 0 if less than 10% of the cells are stained at the 1+ or
higher levels.
Cell 0
Cell 1+
Cell 2+
Cell3+
A couple of interesting observations can be made from the examples above: cell 1+ and cell 2+ show that
cells with and without nuclei are detected, cell 1+ and cell 2+ show that cells with the same membrane
stain intensity can be classified differently based on the membrane completeness, and cell 2+ and cell 3+
show that membranes adjacent to a nucleus as well as membranes a distance from the nucleus are
detected.
Pathologists perform a complex analysis of stained cell features when scoring IHC slides. Aperio’s IHC
Image Analysis algorithms are designed to process the same cell features (nuclei and membrane) and
follow the same scoring schemes as the pathologists when assessing an IHC slide under a microscope.
Being able to do this requires high quality digital slides taken at least with a 20X objective (ScanScopes
provide 20X and 40X).
As with any image analysis algorithm, the IHC Membrane Image Analysis algorithm must be set up for
its specific application by tuning its input parameters. Specific applications may vary in the tissue type,
staining process and/or scoring standard. Cell feature detection parameters and scoring scheme
parameters are handled separately. The cell feature detection parameters specify cell feature detection
thresholds and methods as well as size and shape constraints of cells to distinguish tumor cells from
normal, lymphocyte and stroma cells. The scoring scheme parameters specify the staining intensity and
membrane completeness thresholds that determine the individual tumor cell classification. The scoring
scheme parameters can be trained automatically by presenting the algorithm with a set of digital slides
that have been scored according to a scoring standard (for example: by a panel of pathologists or
longitudinal studies with clinical outcome).
The IHC Membrane Image Analysis algorithm is typically used from within ImageScope, but can also be
used from Spectrum Information Manager or TMALab (Tissue Micro Array). Once a pathologist has
outlined the tumor-cell only regions and runs the algorithm, the algorithm provides a markup image and
an annotation window as its outputs.
IHC Membrane Image Analysis User’s Guide
3
Chapter 1 – Overview
The markup image highlights the detected cell features (black = nuclei and membrane) and the
membrane staining which is color-coded according to the cell classification (blue = 0, yellow = 1+, orange
= 2+, red = 3+). The annotation window provides the HER2 score, the percentages of 0, 1+, 2+ and 3+ cells
and other image analysis statistics. The other image analysis statistics include: average membrane
intensity as intensity value 0 to 255, percentage and number of cells with complete membrane staining,
number of 3+, 2+, 1+ and 0 cells and total number of cells. Providing not only the very coarse HER2 score
but also the percentages of 3+, 2+, and 1+ cells gives the pathologist a very detailed assessment of the slide
and makes it easy for him/her to identify and assess border line cases.
The performance of the IHC Membrane Image Analysis algorithm should be validated following
standard laboratory practices for the specific applications before being used for analysis.
The pathologist using the IHC Membrane Image Analysis algorithm should verify its proper operation on
each slide analyzed.
Aperio will continue to expand and improve our suite of IHC Image Analysis algorithms, striving to
provide you with the tools you need for the standardization and automation of IHC. Please provide us
with your feedback on how we can improve our products to better meet your needs (see “Contact
Information” on page ii for information on contacting Aperio).
4
IHC Membrane Image Analysis User’s Guide
2
Analysis Step-by-Step
This chapter provides step-by-step instructions on how a pathologist would use
the IHC Membrane Image Analysis algorithm as an aid for the semi-quantitative
assessment of an IHC HER2 (Human Epidermal growth factor Receptor 2) breast
tissue slide when reviewing a case via Spectrum Information Manager (SIM).
Access a Digital Slide in SIM (Spectrum Information Manager)
Cases, specimens and digital slides are managed using Aperio’s SIM (Spectrum Information Manager). A
pathologist who wants to access a digital slide first needs to log into SIM and navigate to the case and the
specimen that shows the list of its associated digital slides (H&E, HER2, ER, PR, …).
1.
Please ask your SIM administrator for the SIM web-address, your login name, and a password.
2.
Open Microsoft Internet Explorer (or another web viewer like Firefox) and enter the SIM webaddress. SIM will provide you with a login page.
3.
Log into SIM by entering your login and password and clicking the Login button.
4.
Navigate to the list of cases by clicking List all Cases.
5.
Open the case you want to review by clicking on the open data icon
6.
Open the specimen digital slides list by clicking on the open data icon
specimen.
for the case.
for the case
Patient Data Confidentiality
Spectrum Information Manager provides a full security system that ensures that users can only access
data they are authorized to see, thus protecting patient data from unauthorized access.
Digital slides and related meta-data reside in the Spectrum Information Manager database. Spectrum
Information Manager uses an administrator/user structure that restricts security-sensitive actions to an
authorized administrator. The administrator can:
„
Set up users with passwords. Users must log into Spectrum Information Manager with their user
name and password and be authenticated before they can access data.
„
Set up data groups that restrict access to specific users.
„
Set users’ access permissions to grant them full access, read-only access or no access to the
defined data groups.
IHC Membrane Image Analysis User’s Guide
5
Chapter 2 – Analysis Step-by-Step
Manual Read of a Slide
Traditionally, the pathologist reads the IHC glass slide manually using a microscope. Aperio provides a
digital pathology workflow in which the pathologist can do a manual read of the digital slide.
A good monitor is a recommended investment for any pathologist who uses a computer monitor for the
manual read of a digital slide.
We recommend any high quality LCD monitor meeting the following minimum requirements:
Display Type:
LCD
Screen Resolution:
1680(h) x 1050(v) pixels
Screen Size:
24”
Color Depth:
24 bit
Brightness:
300 cd/m2
Contrast Ratio:
500:1
Stand-alone image analysis systems that work as an adjunct to manual microscopy create extra work for
the lab technicians and the pathologists. Aperio’s IHC Image Analysis algorithms are well integrated into
the Digital Pathology workflow provided by Aperio’s ScanScope systems and Spectrum software suite.
With the touch of a button, the IHC Image Analysis algorithm can be used while you view the digital
slide.
The pathologist views the IHC digital slide in ImageScope. The pathologist might want to view the
corresponding H&E slide as well to determine the tumor regions. ImageScope allows viewing the H&E
and IHC digital slides simultaneously.
6
7.
Open the digital slide in ImageScope by clicking on its thumbnail
digital slide list. ImageScope will open and display the digital slide.
8.
Use the zooming and panning features of ImageScope to view the digital slide.
in the specimen
IHC Membrane Image Analysis User’s Guide
Chapter 2 – Analysis Step-by-Step
Here is a short list of common zooming and panning features in ImageScope (commands in italic are
preferred for fast, single-hand operation):
Zoom:
Adjust zoom level by clicking on the controls in the “zoom window”—fit, 1X, … 20X,
40X—slider.
Increase and decrease zoom by rolling the mouse wheel: toward you (zoom in) and away from you
(zoom out). Click first anywhere on the digital slide to enable this mode. If you accidentally click the
wheel, you start a panning mode not described here – just click anywhere on the digital slide to
return to zoom mode.
Go to maximum zoom level by double-clicking any location of the digital slide.
Pan:
Position your view by clicking on the rectangle (current view) in the slide thumbnail window,
holding and moving to the desired position.
Move your view by clicking on the slide, holding, and moving
Move to an adjacent view by pressing the Shift and arrow-up, -down, -left, and -right keys.
Slide Quality Control
Note that the quality of the manual read of a slide and the IHC Image Analysis depends on the laboratory
following the quality control instructions recommended in the labeling of IHC assays.
The staining process can yield artifacts that need to be assessed properly by the pathologists. Typical
artifacts to consider are: normal epithelium staining, heterogeneous staining, background staining,
cystoplasmic staining as well as edge, thermal and crush artifacts. Staining artifacts should be considered
carefully and regions with staining artifacts may need to be excluded from the manual read and IHC
Image Analysis.
The IHC Membrane Image Analysis algorithm is sensitive to excessive cytoplasmic staining and tissue
being out of focus. It is the responsibility of the pathologist to know the limitations of the algorithm and
to make sure that only good quality digital slides are used for the IHC Membrane Image Analysis
algorithm.
If slides have poor quality, it might be necessary to prepare a new slide or to re-scan the slide.
9.
The pathologist assesses the presence of any staining artifacts, in particular cytoplasmic
staining, and requests the preparation of a new slide if excessive cytoplasmic staining is
observed.
10. The pathologist assesses the focus of the digital slide and requests a re-scan of the slide if the
tissue in the tumor regions is out of focus.
It is recommended that the pathologist follows the appropriate instructions in the specific IHC assay and
associated scoring guide to perform the manual read of the slide.
11. The pathologist determines the HER2 score based on his/her manual read of the slide.
IHC Membrane Image Analysis User’s Guide
7
Chapter 2 – Analysis Step-by-Step
Image Analysis in ImageScope
There are three simple steps involved in performing the IHC Membrane Image Analysis on a digital slide:
(1) outline tumor regions, (2) run algorithm, and (3) view results.
Typically, a pathologist performs all three steps in ImageScope after he has done the manual read of the
slide. To accommodate different workflows, this process can be broken into separate distinct steps in
which the algorithm can be executed from Spectrum Information Manager (batch processing).
Outline Tumor Regions
It is up to the pathologist to identify the tumor regions and outline a representative set of tumor-cell only
regions for the algorithm to analyze. The pathologist needs to make sure he properly identifies invasive
cancer (vs. carcinoma in situ) and does not include regions with staining artifacts in the outlined tumor
regions. For the selection of the tumor regions it is important to appropriately (same ratio of tumor cells)
represent the heterogeneity of staining of a tumor by providing a sufficiently large number of tumor cells
(if possible between 15 and 20 regions and a minimum of about 1000 tumor cells) to maximize robust
analysis results. The IHC Membrane Image Analysis algorithm provides only a limited capability to
identify tumor cells (vs. normal, lymphocyte or stroma cells); therefore the pathologist should only
outline tumor-cell only regions in the tumor regions.
12. The pathologist uses the mouse with the pen tools in ImageScope to outline a representative
and extensive set of tumor-cell only regions to be analyzed by the IHC Membrane Image
Analysis algorithm. Regions can be outlined anywhere in the slide and at any magnification
level (images are processed at 20X).
Here is the list of pen tools in ImageScope (commands written in italic are preferred for fast, single-hand
operation):
8
Pen:
Allows outlining a region that will be processed, activated by clicking on the
pen icon or right-click the mouse and select Pen Tool from the context menu.
Negative Pen:
Allows outlining a region that is excluded from the processing, activated by
clicking on the negative pen icon or right-click the mouse and select Negative
Pen Tool from the context menu. Outlines drawn with the negative pen tool
are shown with a dotted line.
IHC Membrane Image Analysis User’s Guide
Chapter 2 – Analysis Step-by-Step
Note that the pen tools are not persistent—for each region you want to outline you need to click first on
the pen tool of your choice and then outline the region.
Note that if you interrupt the drawing of a region, ImageScope will close the region by connecting the
end points (not visible). If this is not what you wanted, then you should delete the current outlined region
by clicking on the Delete key and starting over again with this region.
To delete any and not only the last region, you need to open the Annotations window by selecting
Annotations Window in the View drop-down menu, select Layer 1 in the Layers area by clicking Layer 1
and selecting the region you want to delete in the Layer Regions area by clicking on the region number.
The selected region will be displayed by ImageScope in the middle of the digital slide view with a black
line in the middle of the colored region outline. To delete the selected region, click on the red X icon
in
the Layer Regions area.
The entire slide is important for the IHC Image Analysis. A prior selection of fields for the image analysis
limits introduces bias to the analysis. Aperio’s IHC Image Analysis algorithms do whole slide processing.
The entire digital slide is available for IHC Image Analysis.
Run Algorithm
The pathologist uses the IHC Membrane Image Analysis algorithm as a macro, which means that none of
the input parameters can be modified by the user. The input parameters have been locked down after
successful setup of the algorithm.
The IHC Membrane Image Analysis algorithm should only be applied for the specific application it is set
up for. Using an algorithm for a different application, like a different tissue type or including stroma and
lymphocyte cells in the tumor regions, can yield meaningless results.
IHC Membrane Image Analysis User’s Guide
9
Chapter 2 – Analysis Step-by-Step
13. To initiate an analysis job, the pathologist selects the Algorithm Server Job window by going
to the View menu and selecting Analysis.
14. The pathologist selects the appropriate algorithm macro for this specific application by
clicking on the macro name.
A lab can create multiple macros of the same IHC Membrane Image Analysis algorithm for different
applications (for example for HER2 reagents from different vendors like Dako and Ventana).
15. The pathologist chooses if the algorithm should generate a markup image by selecting the
Generate Markup Image check box (default: checked).
16. The pathologist starts the execution of the algorithm by clicking the Analyze button.
Digital slides are large images and the analysis of large images requires considerable processing
resources. Lack of resources can create problems such as a time delay or lag in the viewing of the digital
slides by the pathologist if the images are analyzed on his/her computer. Aperio provides “server-side”
processing for its IHC Image Analysis algorithms so that the images are processed where they reside,
thereby eliminating the need for moving the images over the network and using the processing resources
on the pathologist’s computer. Aperio’s “server-side” analysis also provides a scaleable computer-farm
solution that allows easily adding computer resources to the server to meet expanding analysis needs.
10
IHC Membrane Image Analysis User’s Guide
Chapter 2 – Analysis Step-by-Step
View Results
After the analysis is complete, results of the algorithm analysis can be reviewed in the Annotations
window.
17. The pathologist views the Annotations window by selecting Annotation Window in the View
drop-down menu.
The processing of the image is completed when a new layer (Layer 2 in the illustration above) appears as
a child layer to Layer 1 in the Layers area. Layer 1 contains the tumor region outlines provided by the
pathologist. Layer 2 contains the algorithm results. To view the algorithm results, select Layer 2.
18. The pathologist selects Layer 2 by clicking Layer 2 in the Layers area.
The IHC Membrane Image Analysis algorithm provides the following results:
IHC Membrane Image Analysis User’s Guide
11
Chapter 2 – Analysis Step-by-Step
(1) Slide’s HER2 score and percentages of 0, 1+, 2+ and 3+ cells as well as other image analysis statistics
(based on all the cells in all the outlined tumor regions) displayed in the Layer Attributes area.
(2) Regions’ HER2 score and percentages of 0, 1+, 2+ and 3+ cells as well as other image analysis
statistics for each outlined tumor region are individually displayed in the Layer Regions area.
The other image analysis statistics include: average membrane intensity as intensity value 0 to 255,
percentage and number of cells with complete membrane staining, number of 3+, 2+, 1+ and 0 cells and
total number of cells.
(1) Markup Image
The pathologist can view the image of the digital slides with or without the markup image.
The markup image highlights the detected cell features (black = nuclei and membrane) and the
membrane staining which is color-coded according to the cell classification (blue = 0, yellow = 1+, orange
= 2+, red = 3+).
19. The pathologist can view a specific outlined tumor region that corresponds to a region in the
Layer Regions area by clicking on the Region number
.
20. The pathologist views the digital slide with or without the Markup Image by clicking on the
eye icon
(without)
(with markup image) in the Layers area.
Algorithm Quality Control
IHC image analysis is an aid to the pathologists. It is the responsibility of the pathologist to verify the
proper operation of the algorithm.
It is important that the pathologists feel comfortable with the system but identify when the system might
have failed (quality control). Aperio’s IHC Image Analysis algorithms make it easy for the pathologists to
follow what the algorithms are doing and how they determined the scores. By providing meaningful
output of the different processing steps to which the pathologist can relate intuitively, the same cell
features and same scoring scheme used for manual scoring, specifically: (1) a markup image of the
detected cell features and classification of each cell (0, 1+, 2+ and 3+) and (2) the calculated percentages of
cells per cell class from which the scores are calculated directly.
The pathologist should review the markup image for each of the outlined tumor regions and make sure
that the algorithm properly detected and classified the membrane staining for the tumor cells.
12
IHC Membrane Image Analysis User’s Guide
Chapter 2 – Analysis Step-by-Step
21. The pathologist reviews the markup images of the outlined tumor regions and verifies the
proper operation of the IHC Membrane Image Analysis algorithm.
If there are regions where the algorithm did not perform properly, then those regions need to be excluded
from the analysis. To delete regions from the analysis those regions need to be deleted from Layer 1 and
the analysis needs to be re-run on Layer 1 creating a new child layer with the Image Analysis results. You
may want to delete the old Image Analysis result layer Layer 2 before re-running the analysis. Deleting
regions may require adding new regions to make sure that the set of tumor regions used for the analysis
is representative for the tumor. If it is impossible to outline a representative set of tumor regions where
the algorithm performs properly, then the IHC Membrane Image Analysis should not be used for this
particular slide. Please let us know if you have slides on which the IHC Membrane Image Analysis
unexpectedly does not work.
Final Score
The IHC Membrane Image Analysis is intended to be used as an aid to a pathologist. It is the
responsibility of the pathologist to provide the final score based on his/her qualitative assessment and the
quantitative information provided by the IHC Membrane Image Analysis algorithm.
22. The pathologist determines the final HER2 score.
New scoring schemes like the H-Score have been introduced that are based on the same calculated
percentages of cells per cell class as the traditional IHC scores, but provide a finer quantization. Aperio’s
IHC Image Analysis algorithms provide those calculated percentages of cells per cell class as well and
thereby support new scoring schemes including the H-score.
23. When you are done with ImageScope, please exit (go to the File menu and select Exit).
24. When you are done with reviewing cases in SIM, please log out from SIM (go to the Logout
menu) and exit Microsoft Internet Explorer (go the File menu and select Exit).
More Information
The IHC Membrane Image Analysis algorithm is typically used from within ImageScope, but can also be
used from Spectrum Information Manager or TMALab (Tissue Micro Array).
For more details on how to run Aperio algorithms, see:
„
Spectrum Information Manager Operator’s Guide.
„
Spectrum Information Manager Administrator’s Guide.
„
ImageScope User’s Guide.
IHC Membrane Image Analysis User’s Guide
13
Chapter 2 – Analysis Step-by-Step
14
IHC Membrane Image Analysis User’s Guide
3
Installation
This chapter provides instructions on how to install the IHC Membrane Image
Analysis algorithm.
Installing an algorithm is a simple matter of using Windows Explorer to navigate to the location of the
algorithm installer file and double-clicking the file name. This file may be given to you on a CD or your
network administrator may have placed it on your network.
You need to install the algorithm on your ImageServer and locally on the desktops of all users who want
to be able to modify the algorithm parameters and set up the algorithm. Make sure you have installed the
same version of the algorithm on the server and on all desktops.
Follow the instructions of the algorithm installer. After starting the algorithm installer, you see the
welcome window:
1.
On the main installer screen, click Next. A window displaying the Aperio user license appears.
IHC Membrane Image Analysis User’s Guide
15
Chapter 3 – Installation
2.
After reviewing the license carefully, select I accept the terms of the license agreement and click
Next. The Choose Destination Location window appears:
If the location displayed is not where you wish to install the algorithm, click Change to choose a
new location and then click Next; otherwise, just click Next.
3.
On the Ready to Install the Program window, click the Install button. As the installation
proceeds, a status bar shows installation progress. When the installation has completed, the final
window appears. If this window asks if you want to reboot your computer, select No, skip this
step.
4.
Click Finish to exit the installer.
If you upgrade an algorithm to a newer version, the algorithm installer will start with a different
welcome page that provides you with three options: Modify, Repair and Remove. Select Repair and
follow the instructions of the algorithm installer.
16
IHC Membrane Image Analysis User’s Guide
4
Slide Preparation
Slide preparation has a considerable influence on how well image analysis
algorithms will perform. Optimizing the slide preparation for image analysis is a
good idea to get the most out of automatic image analysis. Keep in mind—an
algorithm can only be as good as its input data, the slide. Make sure to control
the quality of the glass slides!
Tissue Preparation
The tissue preparation is important for scan quality. Folds in the tissue and tissue sections that are too
thick will result in blurry images. If working with a blurry slide, preparation of a new slide might be
required or the regions affected need to be excluded from the analysis.
Mechanics
The mechanics of the slide are important for its scan quality and ease of scanning. The slides should be in
clean and good condition—no air pockets under cover slip, no dirt, no fingerprints, no markings, no
writing, no extra adhesive, no broken slides, no chips, no scratches, no overhanging cover slip, etc. The
tissue ideally should be located in the middle of the slide a distance from the edges of the slide, the label
and any other markings. It is helpful for the tissue to be placed consistently in the same location and
orientation on the slide. Some of the mechanical problems of a slide can be resolved by cleaning the slide
with a cotton tissue or trimming the sides with a razor blade. Permanent problems with a slide may
require the preparation of a new slide or excluding the regions affected from the analysis.
IHC Membrane Image Analysis User’s Guide
17
Chapter 4 –Slide Preparation
Staining
Reproducibility of the IHC stains is important for consistent algorithm performance. The algorithm will
be set up for a specific staining process, but it is important to make sure that the variations of the staining
process are controlled and eliminated to the greatest extent possible. For this reason, we recommend the
use of FDA cleared and approved IHC kits and to employ appropriate morphological studies and
controls as specified in the instructions for the IHC kits.
Glass Slide Quality Control
It is the responsibility of the lab to verify the quality of the tissue preparation, the mechanics of a slide
and the staining.
18
IHC Membrane Image Analysis User’s Guide
5
Scanning
The imaging system and its calibration as well as the scanning parameters are
critical to transforming a glass slide into a high quality digital slide. Scan slides
automatically with “one-touch” operation. Make sure to control the quality of
the digital slides!
Imaging System
The performance of any image analysis system is very dependent on the quality of the images it
processes. Aperio’s ScanScope systems are based on Aperio’s patented line-scanning technology which
provides superior high-quality, high-resolution digital slides with virtually no artifacts affecting image
analysis.
Calibration
Color and intensity reproducibility is important for reliable image analysis. The ScanScope systems
provide a high degree of color and intensity reproducibility scan after scan and also from one ScanScope
to another. The ScanScope calibrates each slide when it is scanned by performing a white balance (called
a “prescan”). This is optimum, as the calibration takes into account the characteristics of each slide and
the system at the exact moment of the scan and not at the moment in time when a daily or factory
calibration was performed. To assure proper operation of the ScanScope for each scan, the ScanScope
analyzes the “prescan” of each slide to determine possible variations of the system, like changing
characteristics of the light bulb.
Compression
Digital Pathology would not be practical without a lossy data compression of the large digital slide
images. With Aperio’s ScanScopes, the user can choose between the standard JPEG and JPEG2000 image
compression types and quality factors between 0 and 100.
The IHC Membrane Image Analysis algorithm has been designed to be robust with regard to image
compression artifacts. Nevertheless, you should decide on one compression type and one quality factor
that will be used for the scanning of all IHC slides.
We recommend compressing the images of IHC slides with the standard JPEG2000 compression type and
a quality factor of 70.
Aperio’s ScanScopes provide the capability of applying a sharpening filter while the slides are scanned to
provide pathologists with enhanced sharpness of the digital slides. Image analysis algorithms work best
on raw imagery; therefore the sharpening filter should be disabled during scanning if the digital slides
are intended to be used for image analysis.
IHC Membrane Image Analysis User’s Guide
19
Chapter 5 – Scanning
Resolution for IHC Slides
Pathologists using a standard microscope typically view IHC slides with a 20X objective which is
required to view all relevant cell features for their IHC scoring. The ScanScopes provide slide scanning
with 20X and 40X objectives.
The IHC Membrane Image Analysis algorithm is analyzing the same cell features as assessed by the
pathologists and therefore requires the same image resolution as used by the pathologists. Although
scanning with a 40X objective provides more details, it also requires longer scan times and larger file
sizes.
We recommend scanning IHC slides with a 20X objective.
Scanning Parameter Setting
It is important that all IHC slides are scanned with the same parameter setting. An IHC scanning
parameter set can be set up in the controller XML file to help the operator to make sure that all IHC slides
are scanned with the same parameter settings.
Automatic “One-Touch” Scanning
Aperio’s ScanScopes can scan slides automatically using “one-touch” operation. The ScanScopes provide
a quality factor for each scan from 0 (worst) to 100 (best). Note that a poor quality factor can be caused by
poor slide preparation, like dirt, but that the scan of the tissue still might be good—therefore the operator
should not rely on the quality factor alone to determine the quality of a scan; he/she may want to use the
quality factor as a triage tool to determine which slide he/she wants to review. Poor slide preparation
leads to poor image quality which in some cases can be improved by manually scanning the slide; this
allows the operator to select the tissue region to be scanned, position the “pre-scan” and position and
manually focus the focus points.
Digital Slide Quality Control
It is the responsibility of the ScanScope operator to verify that the entire tissue has been scanned and that
the entire tissue is in focus.
20
IHC Membrane Image Analysis User’s Guide
6
Algorithm Setup
The IHC Membrane Image Analysis algorithm needs to be set up for its specific
application by tuning its input parameters. Specific applications may have
different tissue types, staining processes and/or scoring standards. In the
setup, we distinguish between cell feature detection parameters and scoring
scheme parameters.
Parameter tuning must be done by a skilled user knowledgeable in image analysis and the biochemistry
of the clinical application. If you want help with the setup of the algorithm, please contact Aperio for
image analysis services.
Labs use different IHC reagents and kits (for example: Dako and Ventana) with different staining
characteristics. Different labs also have slightly different IHC staining processes with different staining
characteristics. Aperio’s Image Analysis algorithms provide an automated algorithm training feature that
optimizes the algorithm parameters for specific staining characteristics and scoring standards. Keep in
mind that to assure the accuracy of the IHC Image Analysis algorithms, the lab needs to follow the
quality control instructions recommended in the manufacturer’s labeling of the IHC reagent or kit.
If changes are made in the staining procedure, repeat tuning and validation of the algorithm may be
necessary.
Data Sets
The tuning of the algorithm should be based on a training data set that is representative for the specific
application you want to set up the algorithm for—tissue type, staining process and scoring standard. We
recommend using a minimum of twenty slides with about equal distribution in the HER2 score classes: 0,
1+, 2+ and 3+.
Scoring Standard
As there is no gold standard for IHC scoring, there are different ways you might want to define a scoring
standard for the setup of your algorithm: (1) scores from a panel of three pathologists, (2) a consensus
score by a panel of pathologists or (3) a score based on longitudinal studies relating to the clinical
outcome.
The following shows an example of how to set up the algorithm with a scoring standard based on three
pathologists—the most complex in terms of setting up the training data set, as three pathologists need to
score the same slides independently from each other. Scoring standard training requires both a manual
read of the glass slide as well as a read of the digital slide.
IHC Membrane Image Analysis User’s Guide
21
Chapter 6 – Algorithm Setup
Identify Slides for the Training Data Set
First, you need to identify slides that you would like to use for the training data set. For example, in a
clinical environment where pathologists review digital slides routinely, pathologists could identify
appropriate glass slide by the label and digital slides by their Filename and Date. This information is
easily accessible in ImageScope when viewing a digital slide by selecting Information… in the Image
drop-down menu or by clicking on the information icon . Filename and Date are provided on the
Information tab and the label under the Label Image tab.
Manual Read of Glass Slides
The digital slide scores and image analysis results need to be related back to manual microscopy scores as
those are considered the standard for IHC scoring. Have the three pathologists independently score and
record their reads of the corresponding glass slides using a microscope.
Gather Digital Slides for the Training Data Set
Navigate to the directory where the digital slides reside. Typically, the ScanScope creates the digital
slides after scanning directly in a sub-folder that has the Date as folder name in the D:\Images directory
on the ImageServer.
To separate the training data set from the clinical operation, you need to create a copy of the digital slides.
If you use multiple pathologists to provide blind scores for the scoring standard you need to create a
copy of the training data set for each pathologist—in our example, we therefore need three copies.
To organize the digital slides for the training data set you could create subfolders in the D:\Images
directory. You could have one sub-folder that identifies the specific application and identifies in the
folder name the stain, the manufacturer and tissue type (e.g., HER2_Dako_Breast,
HER2_Ventana_Breast). In this subfolder you could then have a separate subfolder for each copy of the
training data set (e.g., Training1, Training2, Training3).
Organize Training Data Sets in Spectrum Information Manager and Manage Access
Rights
The administrator (or any user with administrative rights) first makes sure that restricted access rights are
set up to protect all data related to the setup of the algorithm and to limit the access to specific data sets to
individual pathologists. Spectrum Information Manager uses a three-step process to set up access rights:
(1) create data groups, (2) grant individual users access to data groups and, (3) associate data to data
groups.
We need one data group for each specific application of the algorithm and individual data groups for
each participating pathologist. Most likely the data groups for the pathologists have already been created
if they are users of Spectrum Information Manager. The administrator logs into Spectrum Information
Manager and navigates to the Data Groups view by clicking Data Groups in the Administrative dropdown menu. The administrator can create a new data group by clicking Create Group in the Data Groups
view and providing a Data Group Name and Data Group Description. For a specific application data
group, use a Data Group Name that identifies the stain, the manufacturer and tissue type and use
Algorithm Setup and Validation as Data Group Description.
22
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
Spectrum Information Manager can grant each user specific access rights to any data group. All
participating pathologists should be users in Spectrum Information Manager and have full access rights
to the specific application data group and their individual data group. Once the setup and/or validation
has been completed, access rights to the specific application data groups can be restricted to Read Only or
No Access to make sure that no data can be modified. If you want other pathologists or customers to be
able to view the data used for the setup and/or validation you can provide them Read Only access rights.
The administrator navigates to the Users view by clicking Users in the Administrative drop-down menu.
To set up a new user, click Create User in the Users view and provide a Login and Password. When you
click Edit for a user, the user view will list all defined data groups. The administrator sets the access level
for each of the pathologists to Full Control for their own data group and the data group for the setup of
the specific application they are participating in. Save settings when done.
IHC Membrane Image Analysis User’s Guide
23
Chapter 6 – Algorithm Setup
Specific applications and their data sets can be organized as cases in Spectrum Information Manager
using specimens for the data sets (think of project for the “case” and of data set for the “specimen”). The
administrator navigates to the Cases view by clicking All Cases in the Cases drop-down menu. To create
a new case (project) click New. As Name, provide a name that identifies the stain, the manufacturer and
tissue type. If you have different stains, different manufacturers or different tissue types or just want to
use different scoring standards you can create a different case (project) for each of these specific
applications to organize the data sets. To identify to the other pathologists that this is not a real case enter
Algorithm as Patient ID.
Open a Case (project) by clicking on the open data icon
for the case.
To restrict the access to the setup for the specific application, select the appropriate data group for the
specific application in the Data Group window of the Case Information and click Save.
Create a new case specimen (data set) for each copy of the training data set and import the corresponding
digital slides. To create a new specimen (data set) click Add New Specimen under Case Specimens. For
the Accession Number enter a name that identifies that this is a training data set and a data set ID
different for each copy of the data set. Click Add to create the new specimen (data set).
Open a specimen (data set) by clicking on the open data icon
corresponding copy of the training data set as new slides:
for the case specimen. Import the
1.
Click Add New Slide under Specimen Digital Slides
2.
Select Add Digital Slides with Images in server folder and subfolders: and provide the folder
name where the copy of the training data set can be found (e.g., D:\Images\
HER2_Dako_Breast\ Training1) and click Add.
To hide any possible patient information from the pathologists viewing the digital slides, you have the
option in the setup of the ImageServer not to show the labels. To randomize the order of the slides in the
training data set, you can add a random number as a pre-fix to the filenames (leave the filename provided
24
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
by the ScanScopes as part of the filename). When Spectrum Information Manager imports digital slides it
does this in alphabetical order and therefore imports the digital slides in the specified random order.
To restrict access to this data set to only the pathologist who should score it, select his/her data group in
the Data Group window of Specimen Information and click Save. Now each pathologist can only see and
access his/her assigned data set.
Manual Read of Digital Slides and Outline of Tumor Regions
Now each of the three pathologists independently reads the digital slides and provides their score and
tumor-cell only region outlines. How a pathologist reviews a digital slide is described in detail in Chapter
2, “Analysis Step-by-Step” on page 5 with two exceptions to this procedure: (1) an annotation file is
imported and the scores are entered by the pathologist so that the scores can be stored in the right format
in Spectrum Information Manager’s data base, and (2) as the algorithms are not set up yet, no image
analysis will be used.
An annotation file needs to be created that has a layer to store the tumor-cell only region outlines and
provides the following Layer Attributes for the scoring in the same layer. It is important to provide the
specified layer attributes by the letter as the algorithm expects exactly those names during the automatic
training of the scoring scheme parameters.
IHC Membrane Image Analysis User’s Guide
25
Chapter 6 – Algorithm Setup
HER2 Score [0,1,2,3]
3+ Percent [0-100%]
2+ Percent [0-100%]
1+ Percent [0-100%]
Please note that in addition to the HER2 score, we also ask the pathologist to estimate the percentage of
3+, 2+ and 1+ cells on which his score was based.
You may want to add other layer attributes of interest to you, like a comment field where the pathologist
could describe staining artifacts he/she observes or the presence of carcinoma in situ, etc.
A master annotation file can be created in ImageScope. The annotation window can be brought up by
clicking Annotation Window in the View drop-down menu. A new layer can be created by clicking on
the new layer icon in the Layers section. The layer can be renamed by clicking on its name, Layer 1. A
new layer attribute can be added by clicking on the add layer attribute icon in the Layer Attributes
section and providing the required attribute names. The master annotations can be exported by clicking
and providing the filename and location.
on the export annotation to file icon
The pathologist can import the master annotation file in ImageScope each time he reviews a digital slide.
He brings up the annotation window by clicking Annotation Window in the View drop-down menu. The
pathologist imports the master annotation by clicking on the import annotation from file icon
and
selecting the appropriate file in the browser. Make sure that the pathologists have file system access to the
master annotation file.
When the pathologist fills out the scores, he/she needs to make sure that he/she only enters numbers, no
% or + signs, and enters a number for each of the layer attributes, even if the number is 0.
When the pathologist has outlined the tumor-cell only regions and provided his/her scores in the
annotation layer attributes, he/she saves his/her annotations in Spectrum Information Manager’s data
base by clicking on the save annotations icon .
Eliminating Outliers
To make sure that the image analysis is in agreement with the manual read of a glass slide, you may want
to exclude digital slides in the training data set when the score provided by the pathologists differs
between their manual read of the glass slide versus the digital slide read.
As there is quite some variability between pathologists scoring an IHC slide, you may want to exclude
digital slides in the training data set where one pathologist does not agree with at least one other
pathologist.
26
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
You can exclude a digital slide from the training data set, either by deleting the digital slide in Spectrum
Information Manager or by excluding the digital slide from the batch analysis. To delete a digital slide in
Spectrum Information Manager, select the check box for the digital slide in the Specimen Digital Slides
list in the specimen (data set) that holds the corresponding copy of the training data set and click Delete.
To exclude a digital slide from batch analysis, unselect the check box for the digital slide when the digital
slides are selected to run the algorithm.
Cell Feature Detection Parameters
The cell feature detection parameters specify cell feature detection thresholds and methods as well as size
and shape constraints of cells to distinguish tumor cells from other cells like normal, lymphocyte and
stroma cells.
The default parameter settings have been optimized for HER2 stains and breast tissue. If you are going to
use the IHC Membrane Algorithm for HER2 stains and breast tissue you may want to first test the default
parameter settings as those may work for you.
Parameter List
It is not necessary nor recommended to modify the following Aperio Algorithm Framework (AAF)
common parameters: View Width = 1000, View Height = 1000, Overlap Size = 100, Image Zoom = 1,
Markup Compression Type = 0 – JPEG and Compression Quality = 30.
Pre-Processing
„
Averaging Radius – Radius (microns) for Noise Reduction [0 < Value < 100] (Default: 1).
The radius of a smoothing filter, which reduces noise resulting in smoother object edges.
Nuclei
„
Threshold Type – Threshold method [0 – Edge Threshold Method, 1 – Manual Threshold]
(Default: 0 - Edge Threshold Method).
A threshold must be applied to the intensity image in order to find the edges of the nuclei.
Threshold type specifies the method to be used to determine the intensity thresholds. Note that
an intensity value of 0 corresponds to black and a value of 255 to white (as bright as possible).
The edge threshold method automatically adjusts the threshold according to the mean of edge
pixels. The algorithm uses an edge finding method to identify edge pixels and uses the average of
these pixel values to determine the threshold. The manual threshold method is the simplest
method, which uses the prescribed intensity thresholds (lower and upper below) to eliminate
unwanted background. This method will not automatically adjust to compensate for lighter or
darker staining between slides.
„
Lower Blue Threshold – Lower limit for nuclear auto thresholding [0 < Value ≤ 255] (Default: 0).
This value can be changed when using the manual threshold type. Since very dark nuclei are
possible, a value of 0 is usually used. Increasing this value will ignore very dark pixels.
„
Upper Blue Threshold – Upper limit for nuclear auto thresholding [0 < Value ≤ 255] (Default:
200).
IHC Membrane Image Analysis User’s Guide
27
Chapter 6 – Algorithm Setup
This value can be changed when using the manual threshold type. A large value (e.g., 240)
represents very faintly stained pixels. Lowering this value to 200 will ignore faintly stained pixels
and require nuclear pixels to have lower intensities (being darker).
„
Blue Curvature Threshold – Curvature threshold for nuclear de-clustering [0 < Value ≤ 25]
(Default: 2.5).
Curvature threshold determines the level of de-clustering used to separate touching nuclei.
Increasing the threshold reduces the effect of the de-clustering logic – this may be necessary if
single nuclei are being split and counted more than once. Decrease this value if more declustering is needed to separate nuclei.
„
Min Nuclear Size – Minimum area for detectable nuclei (micron-squared) [Value > 0] (Default:
15).
Nuclei smaller than this area limit will not be counted.
„
Max Nuclear Size – Maximum area for detectable nuclei (micron-squared) [Value > 0] (Default:
400).
Nuclei larger than this area limit will not be counted.
„
Min Nuclear Roundness – Nuclear objects with roundness less than this are excluded [0 ≤ Value ≤
1] (Default: 0.1).
Roundness is the ratio of the object area to the area of a circle that fully encloses that object.
Circular objects will have a value of 1. Small values indicate non-circular objects.
„
Min Nuclear Compactness – Nuclear objects with compactness less than this area are excluded [0
≤ Value ≤ 1] (Default: 0.1).
Compactness is the ratio of area of the object to the area of a circle that has a circumference equal
to the perimeter of the object. Circular objects will have a value of 1. Small values indicate noncircular objects.
„
Min Nuclear Elongation – Nuclear objects with aspect ratios less than this are excluded [0 ≤ Value
≤ 1] (Default: 0.1).
Elongation is the ratio of the min/max principal moments of the object. Circular objects will have
a value of 1. Small values indicate objects which are long and thin.
Cells/Membrane
„
Cell/Nucleus Requirement – Include/exclude cells without a nucleus [0 – All Cells, 1 – Cells with
Nucleus] (Default: 0 – All Cells).
All cells may not show a nucleus, due to the cutting of the tissue section. If “All Cells” is selected,
then cells are not required to contain a nucleus to be counted. If “Cells With Nucleus” is selected,
then only cells showing a nucleus will be counted.
„
Background Threshold: Membrane intensity threshold for background [0 < Value ≤ 255] (Default:
200).
„
Max Cell Radius – Maximum radius (microns) to limit cell growth [0 < Value < 100] (Default: 5).
Determines cell size for cells that have a nucleus but no membrane staining. A cell is grown
surrounding the nucleus according to this limit.
28
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
„
Min Cell Size – Minimum area for detectable cell (micron-squared) [Value > 0] (Default: 25).
Cells smaller than this area limit will not be counted.
„
Max Cell Size – Maximum area for detectable cell (micron-squared) [Value > 0] (Default: 2000).
Cells larger than this area limit will not be counted.
„
Min Cell Roundness – Membrane (cell) objects with roundness less than this are excluded [0 ≤
Value ≤ 1] (Default: 0.1).
Roundness is the ratio of the object area to the area of a circle that fully encloses that object.
Circular objects will have a value of 1. Small values indicate non-circular objects.
„
Min Cell Compactness – Membrane (cell) objects with compactness less than this area are
excluded [0 ≤ Value ≤ 1] (Default: 0.1).
Compactness is the ratio of area of the object to the area of a circle that has a circumference equal
to the perimeter of the object. Circular objects will have a value of 1. Small values indicate noncircular objects.
„
Min Cell Elongation – Membrane (cell) objects with aspect ratios less than this are excluded [0 ≤
Value ≤ 1] (Default: 0.1).
Elongation is the ratio of the min/max principal moments of the object. Circular objects will have
a value of 1. Small values indicate objects which are long and thin.
Algorithm Modes
The IHC Membrane Image Analysis algorithm has three different modes of operation: (1) Tuning, (2)
Training and (3) Analysis with their respective markup images to choose from. Adjusting the algorithm
parameters is done in tuning mode.
„
Use Mode: Select Training or Analysis/Tuning Mode [0 – Analysis/Tuning, 1- Training] (Default:
0 – Analysis/Tuning).
„
Mark-up Image Type: Choose type of markup image. [0 – Tuning, 1 – Analysis, 2- Analysis (With
Cell Edges)] (Default: 1 – Analysis).
Set up Data Set for Algorithm Parameter Tuning
Digital slides accessed through Spectrum Information Manager can only be analyzed by algorithm
macros after the final parameter settings have been determined. To determine the algorithm parameter
settings, we need to be able to adjust the algorithm parameters, which is only possible when running
algorithms on digital slides that have been opened by ImageScope over the file system (locally or across
the network).
There are different ways to access the digital slides over a file system (local or mapped drive) for the
parameter tuning. In the case that the images folder where the digital slides are stored on the
ImageServer is accessible over the network, those digital slides can be accessed directly by ImageScope
over the file system. If this is not possible, then you need to place a copy of the digital slides somewhere
where you can access them over the file system. Another possibility to create a data set of digital slides
for algorithm tuning is to extract the significant regions from the digital slides (use the annotations from
the pathologists to identify those regions and their significance) and store them locally.
To extract a region in ImageScope:
IHC Membrane Image Analysis User’s Guide
29
Chapter 6 – Algorithm Setup
1.
Click on the Extract Region icon.
2.
Outline the region to extract.
3.
Provide the output file location and filename in the Extract Region Image Region window and
click Extract.
To open a digital slide over the file system with ImageScope, either: Double-click on the file and it will
open with ImageScope or in ImageScope select Open Image… in the File drop-down menu, browse to
the digital slide and click Open.
The annotations from the pathologists provide guidance on which regions should be analyzed and what
their significance is. Because the annotations are stored in Spectrum Information Manager’s database, we
need to export those annotations into the same local folder where the digital slides svs files are stored.
The annotation xml files should be saved with the same filename as their corresponding svs files but with
an xml file extension. When a digital slide svs file and annotation xml file have the same filename,
ImageScope automatically loads the corresponding annotation file when opening the digital slide.
To find the corresponding svs file for a digital slide in Spectrum Information Manager, you can look up
its filename originally provided by the ScanScope in ImageScope’s Image Information window opened by
clicking on the information icon . Note that as the digital slides were imported from a folder, the digital
slides are in inverse order in Spectrum Information Manager as the digital slides in the folder. To save the
annotations to a file, bring up the Annotations window in ImageScope by selecting Annotations Window
from the View drop-down menu, click on the export annotation to file icon
and provide the filename
and location.
When opening digital slides over the file system and running algorithms, the location where the markup
images are stored is determined by ImageScope. If multiple users (multiple ImageScopes) want to see the
markup images, then we need to create a common markup image folder accessible by everybody and
specify this folder in all their ImageScopes. To specify the markup image folder in ImageScope, select
Options… in the Tool drop-down menu, select the General tab and enter the Markup Image Folder under
Analysis.
Also, make sure that the person responsible for the algorithm parameter tuning has the same version of
the algorithm installed on his/her computer as is installed on the ImageServer!
Running the Algorithm and Adjusting the Parameter Settings
Now when the digital slides are opened over the file system in ImageScope, the algorithm parameters can
be adjusted and the results using different parameter settings can be evaluated quickly for a couple of
digital slides.
30
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
1.
First, bring up the Algorithms window by selecting Analysis in the View drop-down menu.
2.
Then select the IHC Membrane Algorithm by clicking Select Algorithm, selecting the IHC
Membrane Algorithm, and clicking Select.
3.
Now the selected algorithm is loaded and its parameters are shown in the Algorithms window
where its settings can be adjusted. Parameters can be modified by changing their values or
selecting a different choice in the drop-down menus. Note that new parameter values only take
effect after the cursor has been moved off the parameter value—once you have modified a
parameter value, just click on a different parameter.
4.
To use the algorithm for parameter tuning, set Use Mode to 0 – Analysis/Tuning. and Markup
Image Type to 0 – Tuning.
When the parameter settings are modified, ImageScope will make them the default parameter
settings as long as the algorithm is loaded. If the same or another algorithm is selected, the
factory default settings are applied and the current parameter settings are lost. To save a
parameter setting you can export the algorithm with its parameter setting as a macro to a file by
clicking Export Macro… in the Algorithms window, providing the file location and filename and
clicking Save.
You may want to save a couple of good parameter settings to be able to evaluate them against
each other over the entire training data set. It might be a good idea to use a filename that
identifies the stain, the manufacturer and the tissue type as well as the Tuning algorithm mode
IHC Membrane Image Analysis User’s Guide
31
Chapter 6 – Algorithm Setup
and a parameter setting identifier. Store the algorithm macros with the training data set so you
know what they belong to and can find them easily.
The user has a choice between Current Screen, the Entire Image and the regions outlined in the
Selected Annotation Layer for the Region of Analysis. Use caution when using Current Screen as
the digital slide will be processed at the resolution defined by the zoom level in ImageScope and
will only be processed at 20X if the zoom level is set to 20X (required for the IHC Membrane
Algorithm). To evaluate different parameter settings, the best approach is to select a small region
that exhibits certain characteristics to which the algorithm should be tuned and select Selected
Annotation Layer and then study the impact of changing the parameter settings. Note that the
processing of the Entire Image takes a long time and does not necessarily provide more valuable
data for the algorithm tuning.
5.
The best way to evaluate the results of an image analysis algorithm is visually; therefore select
the Generate Markup Image check box.
The markup image for tuning shows the nuclei in green with a black outline, the cells outlines or
membranes as black outlines and the membrane staining in red. Note that the scoring scheme
parameters that determine the classification of the cells have not been determined yet therefore
showing the color-coded classification of the cells would not be meaningful.
6.
To run the algorithm, click Run. The algorithm will return its results in a new layer in the
Annotations window. The progress and completion of the analysis is shown at the bottom of the
Algorithms window.
The Annotations window provides a powerful tool for managing the different results provided by the
algorithm for different parameter settings.
For example, we used Layer 1 to outline the regions of analysis and ran the algorithm first with the
default settings and then four times with different values (0, 1, 2 and 5) for the Averaging Radius (um)
parameter creating thereby four child layers with different algorithm results. The algorithm parameters
used to run the algorithm are always stored with the algorithm results under Layer Attributes, making it
32
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
easy to remember the exact parameter settings. Comparing the markup images on the same regions and
outputs from different parameter settings is as easy as clicking on the different layers. To review the
original image, click on the show/hide layer icon in the Layers section of the Annotations window and
toggle between the markup image and the original image.
Test Parameter Settings
To test one or more parameter settings on the entire training data set, import the algorithm macros into
Spectrum Information Manager and run them in batch mode on all digital slides in the training data set
using the annotations provided by the pathologists to define the regions of analysis.
1.
Import an algorithm macro into Spectrum Information Manager by first opening an Internet
browser and logging into Spectrum Information Manager using the same workstation used to
save the tuning parameters and going to the Analysis Macros view by clicking Macros under the
Analysis drop-down menu.
2.
Then in the Analysis Macros view, click Add Macro, provide the tuned parameter set location
and filename and a Macro Name to identify this macro by in Spectrum Information Manager and
click Save. It might be a good idea to use a Macro Name that identifies the stain, the
manufacturer and the tissue type as well as the Tuning algorithm mode and a parameter setting
identifier.
3.
Navigate in Spectrum Information Manager to the training data sets by clicking All Cases under
the Cases drop-down menu and then opening the case (project) for the specific application by
clicking on the corresponding open data icon . Go to the different training data sets by opening
the corresponding specimen (data set) by clicking on the open data icon .
IHC Membrane Image Analysis User’s Guide
33
Chapter 6 – Algorithm Setup
4.
To run an algorithm macro in batch mode on all digital slides grouped under a specimen (data
set) first select all digital slides under Specimen Digital Slides by selecting the check box in the
title bar, and then click Analyze.
5.
In the Analyze Images view, select the algorithm macro in the Select Analysis Macro drop-down
menu, select the annotation layer that contains the annotations from the pathologists with the
tumor-cell only region outlines to be processed in the Select Input Annotation Layer drop-down
menu, and click Analyze.
The Pending Analysis Jobs window then shows the progress on the batch processing. To update
the view, click on the Internet Explorer refresh icon .
6.
To review the results from the different parameter settings, go back to the training data sets, open
all digital slides with ImageScope and compare the markup images of all the regions and
algorithm outputs from the different parameter settings. Determine the best parameter setting.
7.
Delete the parameter tuning algorithm macros in Spectrum Information Manager if they are no
longer required. To delete an algorithm macro, go to the Analysis Macros view by clicking
Macros on the Analysis menu. Then delete a specific macro by clicking Delete after the Macro
Name.
Scoring Scheme Parameters
The scoring scheme parameters specify the staining intensity and membrane completeness thresholds
that determine the individual tumor cell classification. The scoring scheme parameters can be adjusted
34
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
either manually like the cell feature detection parameters or trained automatically by presenting the
algorithm with a set of digital slides that have been scored according to a scoring standard. The automatic
training evaluates all possible threshold combinations to determine the thresholds that provide scores
that best match the established standard scores.
Parameter List
Cell classification thresholds
„
Weak(1+) Threshold: Membrane intensity threshold for weak (1+) [0 < Value ≤ 255] (Default: 200).
„
Moderate(2+) Threshold: Membrane intensity threshold for moderate (2+) [0 < Value ≤ 255]
(Default: 175).
„
Strong(3+) Threshold: Membrane intensity threshold for strong (3+) [0 < Value ≤ 255] (Default:
150).
„
Completeness Threshold: Threshold for classifying membrane as completely stained [0 < Value <
100] (Default: 80).
Algorithm Modes
The IHC Membrane Image Analysis algorithm has three different modes of operation: (1) Tuning, (2)
Training and (3) Analysis with their respective markup images to choose from. Training the scoring
scheme parameters is done in training mode.
„
Use Mode: Select Training or Analysis/Tuning Mode [0 – Analysis/Tuning, 1- Training] (Default:
0 – Analysis/Tuning).
„
Mark-up Image Type: Choose type of markup image. [0 – Tuning, 1 – Analysis, 2- Analysis (With
Cell Edges)] (Default: 1 – Analysis).
Automatic Parameter Training
„
Classifier Type: Select type of classifier [0 – IHCMembrane, 1- IHCNuclear] (Default: 0 –
IHCMembrane).
„
Classifier Definition File: Name of classifier definition file [folder name]. (Default:
IHCMembraneTraining).
Automatic Parameter Training
Create an algorithm macro for the automatic parameter training.
Start with the cell feature detection parameters that yield the best results on the training data sets.
1.
Open ImageScope (just the program) and import the algorithm macro with the best cell feature
detection parameter settings.
2.
To import an algorithm macro, first click Analysis on the View menu to show the Analysis
window.
3.
In the Analysis window, click Import Macro… and provide the file location and filename of the
algorithm macro and click Open.
4.
To use the algorithm for automatic training, set the Use Mode to 1 – Training.
IHC Membrane Image Analysis User’s Guide
35
Chapter 6 – Algorithm Setup
5.
Set the Classifier Definition File parameter to the name of the folder to receive the training
results. The algorithm will create a folder with this name in the ScanScope program folder under
AAF and Algorithms (\Program Files\ScanScope\AAF\Algorithms). As we will run the
algorithm macro on the server, the folder will be created on the ImageServer. Each time you want
to automatically train the scoring scheme parameters, you need to provide a different Classifier
Definition File folder name. You may want to choose a name that identifies the stain, the
manufacturer and tissue type as you need to train the scoring scheme parameters for each
specific application. Note that each time you run the algorithm on a digital slide in training mode
and use an existing Classifier Definition File folder, it will add to the training and update the
trained scoring scheme parameters.
a) To re-start/re-initialize the training, delete the Classifier Definition File folder.
b) To finish the training, change the User Mode to 0 – Analysis/Tuning and the training results
will no longer be updated.
36
6.
Export the algorithm with its parameter setting as a macro to a file. It might be a good idea to use
a filename that identifies the stain, the manufacturer and the tissue type as well as the Training
algorithm mode. Store the algorithm macro with the training data sets so you know where it
belongs and can find it easily.
7.
Import the algorithm macro for automatic training into Spectrum Information Manager. It might
be a good idea to use a Macro Name that identifies the stain, the manufacturer and the tissue type
as well as the Training algorithm mode to identify this macro by in Spectrum Information
Manager.
8.
Run the automatic training algorithm macro on all digital slides in the training data sets except
those that you want to exclude from the training data set. Note that there is no need to create a
markup image for algorithm training as the scoring scheme parameters that determine the
classification of the cells have not been determined yet and the markup images for the cell feature
detection parameters have already been generated. So you may not want to select the option to
create a markup image when running the algorithm macro as this accelerates the processing.
9.
Delete the automatic training algorithm macro in Spectrum Information Manager when the
training is done to ensure that no further digital slides are added to the training by accident and
your training results are preserved.
IHC Membrane Image Analysis User’s Guide
Chapter 6 – Algorithm Setup
10. Retrieve the trained scoring scheme parameters provided by the automatic training. The trained
scoring scheme parameters are provided in a file with the extension .trr in the folder defined by
the Classifier Definition File in the ScanScope program folder under AAF and Algorithms
(\Program Files\ScanScope\AAF\Algorithms\Classifier Definition File). Open the file with
Notepad. The last four numbers in descending order correspond to the cell classification
thresholds. The first three correspond to the intensity thresholds (e.g., Weak (1+) Threshold = 200,
Moderate (2+) Threshold = 175 and Strong (3+) Threshold = 140) and the fourth one to the
completeness threshold (e.g., Completeness Threshold = 60).
IHC Membrane Algorithm
8.001
0
4
2.000000e+002
1.750000e+002
1.400000e+002
6.000000e+001
Create Final Algorithm Macro
Now that you have determined the best parameter settings based on the training data sets, create the final
algorithm macro to be used for the analysis.
1.
Start with the automatic training algorithm macro and set the cell nuclear classification
thresholds to the values provided by the automatic training.
2.
To use the algorithm for analysis, set Use Mode to 0 – Analysis/Tuning. and Markup Image Type
to 1 – Analysis.
3.
Export the algorithm with its parameter setting as a macro to a file. It might be a good idea to use
a filename that identifies the stain, the manufacturer and the tissue type as well as the Analysis
algorithm mode. Store the algorithm macro with the training data sets so you know what it
belongs to and can find it easily.
4.
Import the final algorithm macro for analysis into Spectrum Information Manager. It might be a
good idea to use a Macro Name that identifies the stain, the manufacturer and the tissue type to
identify this macro by in Spectrum Information Manager.
Validate Algorithm Setup
To validate the setup of the algorithm, run the IHC algorithm on the training data sets, then review the
markup images and analyze the scores provided by the algorithm against the scores provided by the
pathologists.
The scores provided by the pathologists and the algorithm can easily be exported from Spectrum
Information Manager into an Excel Spreadsheet. Spectrum Information Manager exports a csv file
(comma separated text file). To open the file directly in Excel, make sure you have Excel installed on your
computer.
IHC Membrane Image Analysis User’s Guide
37
Chapter 6 – Algorithm Setup
1.
To export the annotations layers from Spectrum Information Manager select all digital slides
under Specimen Digital Slides by selecting the check box in the title bar, and then click Export.
You can choose to either Open or Save the file. All the information from all layers and from all
selected digital slides is exported.
In Excel you can delete or hide the columns you are not interested in, resulting in a summary
table like the following. If you use multiple copies of the training data set, you would export a
summary table for each copy of the training data set.
HER2
Score Pathologist
1
0
1
0
2
2
3
3
HER2 Score
- Algorithm
1
0
1
0
3
2
3
3
Use the summary tables for the manual read of the digital slide and image analysis as well as the
manual read of the glass slide scores to do your statistical analysis to evaluate how good your
agreement is between image analysis and the manual scoring for the training data set.
If you find problems with the markup images or do not achieve good agreement on the training
data set, you need to investigate what the origin of the problem is and re-do the setup of the
algorithm.
Once the algorithm setup has been completed, access rights to the specific application data
groups can be restricted to Read Only or No Access to make sure that no data can be modified. If
you want other pathologists or customers to be able to view the data used for the setup you can
provide them Read Only access rights.
38
IHC Membrane Image Analysis User’s Guide
7
Validation
How well the IHC Membrane Image Analysis algorithm will perform for your
application will depend on your setup of the algorithm and your slide
preparation. Before using the IHC Membrane Image Analysis algorithm for
analysis, we recommend validating the performance of the algorithm.
How to validate the IHC Membrane Image Analysis algorithm for your application is entirely up to you.
We show an example of what we think could be a possible study design and how to use Spectrum
Information Manager to execute the study.
Study Design Example
As there is no gold standard for IHC scoring, we base the validation on the scores from a panel of three
pathologists. The goal for the performance validation study should be to demonstrate substantial
equivalence of IHC Membrane Image Analysis algorithm to conventional manual microscopy. This can
be demonstrated by calculating measures of reliability separately for pathologists using manual
microscopy and the image analysis system, and across both methods. Substantive equivalence will be
established by finding no substantial differences for the separate and combined estimates of reliability.
The lab should use FDA approved kits/reagents for HER2. The lab should employ appropriate
morphological studies and controls as specified in the instructions for the IHC kits/reagents.
The evaluation slides should be taken from the clinical operation as the cases come in. A sufficient
number of slides per HER2 score: 0, 1+, 2+ and 3+ are required to provide sufficient power for the
statistical analysis.
The slides in the data set should pass a quality inspection by a pathologist in terms of having a good
representation of tumor regions present in the specimen and cleanness and good condition of the slides in
order to be included in the validation study.
The slides should be made anonymous and therefore not contain any human subject information.
The study should have three blind evaluation arms: manual microscopy, digital slide scoring and image
analysis.
IHC Membrane Image Analysis User’s Guide
39
Chapter 7 - Validation
Microscope
Pathologist #1, #2, #3
Manual
Microscopy
Slide Score
Glass Slides
Wash-out
Randomize
ScanScope®
ImageScope
Pathologist #1, #2, #3
Digital
Slide
Scoring
Slide Score
Glass Slides
Image
Analysis
Slide Score
Tumor Regions
Annotations
Image Analysis
Manual Microscopy
Three pathologists should score independently the glass slides using manual microscopy.
Between the manual microscopy and the digital slide scoring there should be a wash-out period of at
least one week and the slides should be randomized and scanned.
Digital Slide Scoring
The same three pathologists should then score independently the digital slides and outline the tumor
regions.
Image Analysis
The image analysis should then be performed on all digital slides analyzing them by the pathologists’
outlined tumor regions. We separate the image analysis from the digital slide scoring so that we could
easily re-run the validation study when new algorithm versions are released by just re-running the
algorithm and performing the statistical analysis.
40
IHC Membrane Image Analysis User’s Guide
Chapter 7 - Validation
Validate Algorithm Performance
The performance validation data set can be organized in Spectrum Information Manager exactly the same
way as the training data set for the algorithm setup using the same cases (projects) for the specific
applications.
The only difference is the annotation file that only needs to include the HER2 score.
The rest of the validation is the same as the validation of the algorithm setup.
If you find problems with the markup images or do not achieve good agreement between scores, you
need to investigate what the origin of the problem is and go back to the setup of the algorithm.
Once the algorithm validation has been completed, access rights to the specific application data groups
can be restricted to Read Only or No Access to make sure that no data can be modified. If you want other
pathologists or customers to be able to view the data used for the validation you can provide them Read
Only access rights.
IHC Membrane Image Analysis User’s Guide
41
IHC Membrane Image Analysis User’s Guide
MAN-0026, Revision B
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising